Our objective is to develop anatomical models to use in computer programs that analyze complex medical images. With these programs we hope to augment the specialist's perceptual capabilities. This research will expand our previous work on the location of chest nodules in radiographs and extend these ideas to the problem of the location of heart walls in ultrasound images. On the first problem, we propose to: (1) modify our programs so that they can detect nodules of the 3-5 mm diameter; (2) develop rib models and rib-finding procedures and use them to refine our nodule-locating programs; (3) extend our nodule model; and (4) test our programs on a data base of 100 radiographs, of which at least 60 contain solitary nodules. On the second problem we propose to: (1) find model-based algorithms for detecting the heart wall boundaries in three-dimensional, time-varying ultrasound image data; (2) use boundaries found in different frames to create a surface representation of the heart walls; and (3) implement graphics software to display the time-varying heart wall surfaces. Although these problems arise in different medical specialties and involve different imaging technologies, we believe the computer models of anatomy and basic methodology needed to solve them are similar. The methodology includes multi-resolution image analysis, and image segmentation into surfaces and edges using semantics-based algorithms. The methods developed as part of this research could be used in the computer-aided diagnoses of many other medical images. Studying problems from two different domains raises the chances that the methods will carry over to other domains.